v0.16.0

Changed

• kac export carries a field its type declares as a list. Every such field was written as null on every record,
  which reads exactly as a record that holds nothing. A list now travels as a JSON array, and an entry the type
  declares as an object carries the keys that declaration names. A list a record left empty stays null, beside the
  field it never wrote. docs/design/export.md states the shape.

• The glossary exports tags. It is the first field to travel as a list. A consumer holding a vendored glossary
  can filter its records by subject without reading each Scope. The key lands on glossary/<record>.json, and a term
  line in terms.jsonl carries no tags.

• kac new no longer sends a corpus a link to a type it declined. A type's root page and its _template.md name
  the other types and link to them, which is what makes a full corpus navigable and what left a corpus adopting a subset
  holding dead links. Each page is now unlinked as it is written: a reference to a declined type keeps its own wording
  and loses its link, so That is a [service](services.md). arrives as That is a service.

  The same happens on kac update --add-type, for the page that arrives, and on kac update --policy full, which now
  holds a seed to the template as this corpus would have received it rather than as it was authored. Without that a full
  update wrote the links back.

  This reaches the pages a corpus receives once and then owns. A framework document is shared word for word, and
  framework-names-types goes on holding it to naming a type rather than linking to one.

  A link into another type's folder has no such repair, because its text names a record. Two seed pages defined one as
  a reference link, which reached a corpus whole, so glossary.md and frameworks.md now name the record without
  linking it.

• kac update --add-type says what the arriving page does not get. The new page links to the types the corpus
  holds. The pages already there name it without linking, because each was written while the type was still declined,
  and changing them is the corpus's own call.

• kac validate no longer refuses a schema for naming a type the corpus declined. A field's ref: and a type's
  versus: each name a type, and a corpus adopts as many types as it has use for. .schema/standards.yaml alone
  reaches four other types, through ref: on four fields and a versus: naming one of them again, so a corpus adopting
  standards and nothing else met five schema-dispatch errors on a schema it had just been sent. Both declarations are
  now left alone where no schema covers the type. Nothing is rendered, and kac update --add-type starts the reference
  without an edit to .schema/.

  What a record is held to does not soften with it. ref-resolves goes on asking that a cited id exists, and a field
  whose every declared type this corpus turned down now admits nothing rather than everything. It names the types the
  declaration wanted, since a type nothing covers has no label to read:
  'derived-from' points at 'std-0002', which is a Standard. The field points at 'adrs', which this corpus did not adopt.

  With this and the unlinking above, kac new adopting any single type writes a corpus that validates and exits 0.

• kac update --drop-type asks before it deletes. Giving up a type deletes its page and leaves every page still
  naming it holding a dead link. The run says so, says that kac validate reports the ones it can reach, and waits for
  an answer. The question takes no by default. --yes answers it in advance, and a run with no terminal and no
  --yes refuses rather than guessing.